Week 10 - Introduction to AI and Machine Learning

AI and ML with an activity

October 28, 2022

Activity session 🏫 Date: Oct 28, 2022, Friday Event: Classroom session Venue: KMUTT S4 building 11th Floor Room 1114

Instructors (s):

1. P'North (IG: north.npk) AI researcher at NSTDA

We learned about the terms AI (Artificial Intelligence) and Machine Learning during this week.

Coursework (the 'what?') 🤷‍♂️

Artificial intelligence (AI) is the capacity of a digital computer or computer-controlled robot to carry out operations frequently performed by intelligent beings. This is a broad area of computer science that is focused on creating intelligent machines that can complete jobs. AI allows robots to adapt to new input, learn from past experiences, and carry out activities like to those performed by humans using a dataset. We also learned about deep learning, a form of machine learning and artificial intelligence (AI) that mimics how people learn specific kinds of information.

In the class, we also prepared a word cloud about AI applications, these were some of the uses that caught my attention:

• AI can automate mundane tasks and free up resources for more complex tasks.

• AI can improve decision-making processes by providing data-driven insights.

• AI can analyze large datasets quickly and accurately.

• AI can recognize patterns in data and can use this knowledge to inform decisions.

• AI can process natural language, enabling humans to interact with machines using voice or text.

• AI can help in the development of autonomous vehicles and robotics.

• AI can be used to detect fraud and financial risks.

• AI can help identify health issues and recommend treatments.

These were the potential applications that inspired me to dig deep into AI.

AI and ML are related but distinct fields. AI is a broad field of technology that focuses on creating intelligent machines that can think and make decisions like humans. ML, on the other hand, is a specific subfield of AI that focuses on the development of algorithms and models that can learn from data and make predictions. AI is a more general field that includes ML, as well as other techniques such as natural language processing, computer vision, robotics, and more.

We explored several topics in the session, which can be checked in the bottom part of the reflective writing (if it happens to be your interest).

Impact (the 'so what?') 🚀

We can gain a competitive edge in business by automation, improving customer service, designing a better product, reducing the cost of human errors, and having data-driven insights.

The lab session we had after the instruction part was a stepping stone to exploring AI projects. Before then I had always wanted to learn about AI but entry to this topic was too confusing and vague especially when we don't have mentions. I am eternally grateful to P'North for this wonderful session.

Reflections (the 'now what?') 🤔

AI may be used to create engaging learning activities like project-based learning and gamified learning, among others. And Machine learning (ML), a branch of artificial intelligence (AI), is the branch of computational science that is concerned with the analysis and interpretation of patterns and structures in data to enable learning, reasoning, and decision-making without the involvement of a human.

With these sessions and learning materials provided, now I have a good roadmap and interest to explore more about AI technologies, trends, and applications. Moreover, it helped me get involved firsthand rather than being an audience spectating other players in this technology. I also got to clear my misconception that I had before, regarding the technicalities of AI deployment to the client side.

Further reading (optional) 📄

Keynotes 📝

What is artificial intelligence?

• The concept is to learn from experience (Alan Turing, Turing machine at 1953) on his report published in 1948 (Intelligent Machines)

Evolution of AI

• AI (the 1950s) was the era of Turing's work

• ML (the 1980s) brought things like spam filtering

• DL (the 2010s) brought topics like backpropagation

Machine Learning

• TED Talk by Fei Fei Li (How we're teaching machines to understand pictures?)

• Traditional process: data + rules -> outcome

• ML process: data + outome -> rules

Types of Machine Learning Based on the process: Unsupervised, Semi-Supervised, Supervised, Reinforcement (reward-based)

Based on structural data: Linear regression, KNN (K-nearest neighbor), Decision Tree

For Time Series Data: ARIMA, Long Short Term Memory, Recurrent Neural Network

For Natural Language: Also, long short term memory, Transformer model (too complex and hard to learn and use compared to others)

For image data: Convolutional neural network (CNN), Vision transformer, YOLOv7

Real-world usage examples:

1. Netflix recommendation (clustering model)

2. Falcon 9 landing (gliding system)

3. Tesla (Level 2 Automation) driver's assistance autopilot

4. Mid journey Image Generator

What do you want AI to do? -> question (on mentimeter word cloud)

ML Tools: Programming languages: Python 🌟, Scala, Java R

Models:

• CNN (Convolutional Neural Network)

• ANN (Artificial Neural Network), a part of Deep learning

• GAN (Generative Adversarial Network)

• Transformers model (huge but accurate) - state-of-the-art model nowadays

Tools:

1. Pandas - data preprocessing, data frame, import & cleaning

2. sklearn - models, criteria mind mappinga

3. TF Keras

4. YOLO - Computer Vision, Object Detection

5. Kaggle - Find Datasets & example notebooks

6. Papers with code - latest papers and extra datasets

7. Teachable Machine (easily learning ML)

New ideas 💡

• the above key points are commonly found in general talk about machine learning, but most of the miss the calibration metrics part using confusion matrix (it is whether a model is accurate enough to deploy or not)

• Machine Learning follows Garbage In, Garbage Out workflow. Therefore, the quality of data is very much crucial for proper functioning.

• Imbalance in the datasheet?

  • Solution: more data, sampling method

• Also, I got the opportunity to clear some of the issues I've encountered when exploring machine learning on my own:

  1. The data format of the model (JSON/Binary) - Flat file, memory, python variable? Each ML infrastructure has its own, and wrapper to work with client-side with quality documentation. Therefore, no more hurdles to worry about.

  2. Deploy (JS, Qt, Flutter)? The web is the best option (JS). The flutter is quite broken. The backend and REST approach is quite laggy.

  3. Starting a career!

  4. How to choose a project [hackathon]? (hackathon/technology/fun) work with example projects a lot as a beginner

Resources 🎁

• Papers with Code (state of art ML workflow and papers) + extra datasets

• Kaggle to start the work with AI, DL, ML